The plane Countries, relative to each other
Where each country's news leans · and how sure we are
An AI reads the same story in newsrooms across Europe and North America, compares how each one frames it, and places every outlet on a map — with an oval for how sure it is.
Each dot is a country — the weighted average of its largest news brands — placed by how they frame the same events. Colours are regions; the table under the map has every country.
Keep scrolling: time runs with you. Each dot drags its history behind it like a string.
Pick a region above the table to compare neighbours — here, the Nordics. The board zooms to fit them.
Open a country and its own outlets appear, placed relative to the country's position. The table lists them one by one.
Smaller outlets are measured too, and drawn hollow: they weigh little in their country's position.
The judge reads two articles about the same event — outlet names hidden — and says which frames it further right or more conservatively, and why. Millions of these place the dots.
Pick any country on the map or in the table, or drag the timeline. Hover a dot to follow it.
Every hour BasedNews reads the feeds of the largest news brands in 41 countries, and every day it compares how they report the same events. The years before go back to November 2015, rebuilt from the GDELT archive.
Every country and outlet sits on one scale: open a country and its outlets are exactly where they are on the world view. The centre is the average country today, not a political centre.
What is measured is the framing of the same event, not which stories an outlet chooses to cover. An outlet is drawn once it has 5,000 comparisons behind it, a country once its outlets together have; below that a position is inconclusive and is listed, not placed. So the map fills in as the evidence does. The history was judged by a fine-tuned open model on a local GPU, the live days by an OpenAI model, both with the same prompt and the same update rule.
Prompts, assumptions, experiments and the comparison with outside ratings are public in the repository. Article text is never republished.
Every dot rests on two counts: the articles we hold, and the pairs among them — two outlets reporting the same event — that the judge can compare. Week by week, since November 2015.
An open-source project by Axel Solhall. Code, data pipeline and method: github.com/axelsolhall/BasedNews.
Hosting and the daily judge cost a few dollars a month.